Using W&B beyond experiment tracking

Weights & Biases · Beginner ·📊 Data Analytics & Business Intelligence ·3y ago

Key Takeaways

The video demonstrates the advanced features of Weights & Biases, including Sweeps for hyperparameter optimization, Tables for data visualization, Artifacts for data lineage tracking, and Models for storing and reproducing model development.

Full Transcript

the weights and biases platform has evolved into so much more than just experiment tracking and using all of our tools together makes the ml workflow even more seamless what's nice about weights and biases is no matter what your workflow we integrate with all of the popular Cloud providers Hardware vendors and ml Frameworks and libraries as ml projects grow Beyond just a few people that's where the true power of weights and biases can be felt ways and biases helps teams improve collaboration and productivity by being the central system of record across an organization okay so let me show you some of the other things that you can do at weights and biases Beyond just experiment tracking once you've started tracking your experiments in a centralized location the next step is often to optimize model performance and we have a tool called sweeps that does just that it provides lightweight automated hyper parameter optimization and a fully scalable and customizable way as you can see from this parallel coordinates chart you can quickly draw insights into what parameter combinations are leading to the best models and if I drill down into any of these hyper parameters even further you'll see that the data changes to reflect that the platform also allows you to easily visualize your data in various charts and tables here too here you'll see your weights and biases table this interactive data visualization tool allows teams to easily visualize analyze and debug models collaboratively with tables you can combine rich media like images video and audio alongside performance metrics for analysis it's really easy to manipulate a ways and biases table and derive some valuable insights in this case I'm using the table to create a new plot to explore my model predictions debug issues and simply understand how my data sets models and outputs are related I can explore Tables by dynamically filtering grouping by classes or creating new columns to tell me when my model predictions were incorrect to sum it up weights and biases makes tracking visualizing and sharing your work easy because we offer tools like experiment tracking sweeps and tables all in one unified platform if you need to capture the data you're using and producing during experimentation our next tool artifacts can help artifacts tracks your data lineage throughout a Project's life cycle it allows you to version organize and compare data and model assets over time built on top of artifacts is weights and bias models here you can store models together with the experiments that produce them this allows all work throughout the entire model development process to be easily reproducible and discoverable by anyone on the team it's great for onboarding and off-boarding team members as work is never lost now that you have a better understanding of your experiments and what worked and what didn't it's time for you to share your results with your team I'm going to navigate to reports and here you can see a list of reports from this team's project and I'm going to select the publish report image classification live dashboard you can effectively share key project metrics and Analysis with both Technical and non-technical stakeholders without having to take screenshots or clean up any unorganized notes these dashboards are all live so if I do any more work this will automatically update so that the rest of my team or my manager can see my latest progress with reports I can show how my model Works show graphs and visualizations of how the model versions improved over time discuss bugs and demonstrate progress towards key Milestones you can even share your reports externally if an outside audience needs to consume the results as well and that's all for this short video series but don't let that stop you on your Learning Journey here are some useful links for you to dive deeper into what we covered in this video you can check out our documentation if you're having any issues you can go to our support forum and if you just want to learn what other people are using weights and biases for you can browse our community blog fully connected thanks very much and good luck

Original Description

This lesson covers the advanced features of the Weights & Biases platform, including Sweeps for hyperparameter optimization, interactive Tables for data visualization and evaluation, Artifacts for data lineage tracking, and Models for storing and reproducing model development. The lesson also highlights Reports for sharing key metrics and analysis with stakeholders, and demonstrates how W&B integrates with popular cloud providers, hardware vendors, and ML frameworks. Additional links: - W&B Sweeps: https://docs.wandb.ai/guides/sweeps - W&B Tables: https://docs.wandb.ai/guides/data-vis/tables-quickstart - W&B Artifacts: https://docs.wandb.ai/guides/artifacts/ - W&B Reports: https://docs.wandb.ai/guides/reports) - W&B Integrations: https://docs.wandb.ai/guides/integrations) Get your free W&B101 and more certificates here: https://www.wandb.courses/collections
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Playlist

Uploads from Weights & Biases · Weights & Biases · 0 of 60

← Previous Next →
1 0. What is machine learning?
0. What is machine learning?
Weights & Biases
 2 1. Build Your First Machine Learning Model
1. Build Your First Machine Learning Model
Weights & Biases
 3 Intro to ML: Course Overview
Intro to ML: Course Overview
Weights & Biases
 4 2. Multi-Layer Perceptrons
2. Multi-Layer Perceptrons
Weights & Biases
 5 3. Convolutional Neural Networks
3. Convolutional Neural Networks
Weights & Biases
 6 Weights & Biases at OpenAI
Weights & Biases at OpenAI
Weights & Biases
 7 Why Experiment Tracking is Crucial to OpenAI
Why Experiment Tracking is Crucial to OpenAI
Weights & Biases
 8 4. Autoencoders
4. Autoencoders
Weights & Biases
 9 5. Sentiment Analysis
5. Sentiment Analysis
Weights & Biases
 10 6. Recurrent Neural Networks [RNNs]
6. Recurrent Neural Networks [RNNs]
Weights & Biases
 11 7. Text Generation using LSTMs and GRUs
7. Text Generation using LSTMs and GRUs
Weights & Biases
 12 8. Text Classification Using Convolutional Neural Networks
8. Text Classification Using Convolutional Neural Networks
Weights & Biases
 13 9. Hybrid LSTMs [Long Short-Term Memory]
9. Hybrid LSTMs [Long Short-Term Memory]
Weights & Biases
 14 Toyota Research Institute on Experiment Tracking with Weights & Biases
Toyota Research Institute on Experiment Tracking with Weights & Biases
Weights & Biases
 15 Weights and Biases - Developer Tools for Deep Learning
Weights and Biases - Developer Tools for Deep Learning
Weights & Biases
 16 Introducing Weights & Biases
Introducing Weights & Biases
Weights & Biases
 17 10. Seq2Seq Models
10. Seq2Seq Models
Weights & Biases
 18 11. Transfer Learning for Domain-Specific Image Classification with Small Datasets
11. Transfer Learning for Domain-Specific Image Classification with Small Datasets
Weights & Biases
 19 12. One-shot learning for teaching neural networks to classify objects never seen before
12. One-shot learning for teaching neural networks to classify objects never seen before
Weights & Biases
 20 13. Speech Recognition with Convolutional Neural Networks in Keras/TensorFlow
13. Speech Recognition with Convolutional Neural Networks in Keras/TensorFlow
Weights & Biases
 21 14. Data Augmentation | Keras
14. Data Augmentation | Keras
Weights & Biases
 22 15. Batch Size and Learning Rate in CNNs
15. Batch Size and Learning Rate in CNNs
Weights & Biases
 23 Applied Deep Learning Fellowship Overview and Project Selection with Josh Tobin (2019)
Applied Deep Learning Fellowship Overview and Project Selection with Josh Tobin (2019)
Weights & Biases
 24 Grading Rubric for AI Applications with Sergey Karayev (2019)
Grading Rubric for AI Applications with Sergey Karayev (2019)
Weights & Biases
 25 16. Video Frame Prediction using CNNs and LSTMs (2019)
16. Video Frame Prediction using CNNs and LSTMs (2019)
Weights & Biases
 26 Image to LaTeX - Applied Deep Learning Fellowship (2019)
Image to LaTeX - Applied Deep Learning Fellowship (2019)
Weights & Biases
 27 17. Build and Deploy an Emotion Classifier (2019)
17. Build and Deploy an Emotion Classifier (2019)
Weights & Biases
 28 Applied Deep Learning - Data Management with Josh Tobin (2019)
Applied Deep Learning - Data Management with Josh Tobin (2019)
Weights & Biases
 29 Snorkel: Programming Training Data with Paroma Varma of Stanford University (2019)
Snorkel: Programming Training Data with Paroma Varma of Stanford University (2019)
Weights & Biases
 30 Applied Deep Learning - Troubleshooting and Debugging with Josh Tobin (2019)
Applied Deep Learning - Troubleshooting and Debugging with Josh Tobin (2019)
Weights & Biases
 31 Troubleshooting and Iterating ML Models with Lee Redden (2019)
Troubleshooting and Iterating ML Models with Lee Redden (2019)
Weights & Biases
 32 Designing a Machine Learning Project with Neal Khosla (2019)
Designing a Machine Learning Project with Neal Khosla (2019)
Weights & Biases
 33 Lukas Beiwald on ML Tools and Experiment Management (2019)
Lukas Beiwald on ML Tools and Experiment Management (2019)
Weights & Biases
 34 Building Machine Learning Teams with Josh Tobin (2019)
Building Machine Learning Teams with Josh Tobin (2019)
Weights & Biases
 35 Pieter Abeel on Potential Deep Learning Research Directions (2019)
Pieter Abeel on Potential Deep Learning Research Directions (2019)
Weights & Biases
 36 Testing and Deployment of Deep Learning Models with Josh Tobin (2019)
Testing and Deployment of Deep Learning Models with Josh Tobin (2019)
Weights & Biases
 37 Five Lessons for Team-Oriented Research with Peter Welder (2019)
Five Lessons for Team-Oriented Research with Peter Welder (2019)
Weights & Biases
 38 Applied Deep Learning - Rosanne Liu on AI Research (2019)
Applied Deep Learning - Rosanne Liu on AI Research (2019)
Weights & Biases
 39 Making the Mid-career Leap from Urban Design to Deep Learning/Data Science
Making the Mid-career Leap from Urban Design to Deep Learning/Data Science
Weights & Biases
 40 Organizing ML projects — W&B walkthrough (2020)
Organizing ML projects — W&B walkthrough (2020)
Weights & Biases
 41 Brandon Rohrer — Machine Learning in Production for Robots
Brandon Rohrer — Machine Learning in Production for Robots
Weights & Biases
 42 Nicolas Koumchatzky — Machine Learning in Production for Self-Driving Cars
Nicolas Koumchatzky — Machine Learning in Production for Self-Driving Cars
Weights & Biases
 43 My experiments with Reinforcement Learning with Jariullah Safi
My experiments with Reinforcement Learning with Jariullah Safi
Weights & Biases
 44 Applications of Machine Learning to COVID-19 Research with Isaac Godfried
Applications of Machine Learning to COVID-19 Research with Isaac Godfried
Weights & Biases
 45 Testing Machine Learning Models with Eric Schles
Testing Machine Learning Models with Eric Schles
Weights & Biases
 46 How Linear Algebra is not like Algebra with Charles Frye
How Linear Algebra is not like Algebra with Charles Frye
Weights & Biases
 47 Predicting Protein Structures using Deep Learning with Jonathan King
Predicting Protein Structures using Deep Learning with Jonathan King
Weights & Biases
 48 Rachael Tatman — Conversational AI and Linguistics
Rachael Tatman — Conversational AI and Linguistics
Weights & Biases
 49 Reformer by Han Lee
Reformer by Han Lee
Weights & Biases
 50 Sequence Models with Pujaa Rajan
Sequence Models with Pujaa Rajan
Weights & Biases
 51 GitHub Actions & Machine Learning Workflows with Hamel Husain
GitHub Actions & Machine Learning Workflows with Hamel Husain
Weights & Biases
 52 Look Mom, No Indices! Vector Calculus with the Fréchet Derivative by Charles Frye
Look Mom, No Indices! Vector Calculus with the Fréchet Derivative by Charles Frye
Weights & Biases
 53 Jack Clark — Building Trustworthy AI Systems
Jack Clark — Building Trustworthy AI Systems
Weights & Biases
 54 Surprising Utility of Surprise: Why ML Uses Negative Log Probabilities - Charles Frye
Surprising Utility of Surprise: Why ML Uses Negative Log Probabilities - Charles Frye
Weights & Biases
 55 Track your machine learning experiments locally, with W&B Local - Chris Van Pelt
Track your machine learning experiments locally, with W&B Local - Chris Van Pelt
Weights & Biases
 56 Antipatterns in open source research code with Jariullah Safi
Antipatterns in open source research code with Jariullah Safi
Weights & Biases
 57 Attention for time series forecasting & COVID predictions - Isaac Godfried
Attention for time series forecasting & COVID predictions - Isaac Godfried
Weights & Biases
 58 Made with ML - Goku Mohandas
Made with ML - Goku Mohandas
Weights & Biases
 59 Angela & Danielle — Designing ML Models for Millions of Consumer Robots
Angela & Danielle — Designing ML Models for Millions of Consumer Robots
Weights & Biases
 60 Deep Learning Salon by Weights & Biases
Deep Learning Salon by Weights & Biases
Weights & Biases

The video teaches how to use Weights & Biases beyond experiment tracking, including Sweeps, Tables, Artifacts, and Models, to improve collaboration and productivity in machine learning workflows. It matters because it helps teams work more efficiently and effectively. The video provides a comprehensive overview of the platform's features and how to use them to optimize model performance, visualize data, and track experiments.

Key Takeaways
  1. Start tracking experiments in Weights & Biases
  2. Use Sweeps for hyperparameter optimization
  3. Create interactive Tables for data visualization
  4. Use Artifacts for data lineage tracking
  5. Store models with experiments using Models
  6. Share results with team using Reports
💡 The Weights & Biases platform provides a unified system for tracking, visualizing, and sharing machine learning experiments, making it easier for teams to collaborate and work more efficiently.

Related Reads

📰
Why I Left China as a Data Analyst
A data analyst's decision to leave China due to challenges like tech decoupling and AI barriers, and how it impacted their career growth
Hackernoon
 📰
The Presence Premium: Office Mandates Need a 10% Productivity Miracle to Break Even.
Office mandates come with significant costs, requiring a 10% productivity boost to break even, highlighting the need for data-driven decision making in return-to-office strategies
Medium · Data Science
 📰
From Python Basics to Data Analysis: My 4-Week Internship Journey at Sparks To Ideas
Learn how a B.Tech student applied Python basics to data analysis in a 4-week internship, gaining practical experience and skills in data science
Medium · Data Science
 📰
From Python Basics to Data Analysis: My 4-Week Internship Journey at Sparks To Ideas
Learn how a student applied Python basics to data analysis in a 4-week internship, gaining hands-on experience and skills in a real-world setting
Medium · Python
Up next
How to Scrape Facebook Ad Library Data + Analyse on n8n 🔥
DroidCrunch
Watch →